Performance Modeling and Benchmarking of Event-Based ... - DVS
Performance Modeling and Benchmarking of Event-Based ... - DVS
Performance Modeling and Benchmarking of Event-Based ... - DVS
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
Chapter 7<br />
Conclusions <strong>and</strong> Outlook<br />
With the growing popularity <strong>of</strong> EBS <strong>and</strong> their gradual adoption in mission critical areas, the<br />
need for novel techniques for benchmarking <strong>and</strong> performance modeling <strong>of</strong> EBS is increasing.<br />
Since their reliability is crucial for the whole IT infrastructure, a certain QoS level has to be<br />
ensured. The motivation for this thesis was to support the development <strong>and</strong> maintenance <strong>of</strong><br />
EBS that meet certain QoS requirements. Given that EBS differ from traditional s<strong>of</strong>tware in<br />
fundamental aspects such as their underlying communications paradigm, specific solutions <strong>and</strong><br />
concepts are needed. System architects <strong>and</strong> deployers need tools <strong>and</strong> methodologies, which<br />
allow us to evaluate <strong>and</strong> forecast system performance <strong>and</strong> behavior in certain situations to identify<br />
potential performance problems <strong>and</strong> bottlenecks. Benchmarks <strong>and</strong> performance modeling<br />
techniques are usually the means <strong>of</strong> choice to answer these questions. However, no general performance<br />
modeling methodologies focusing on EBS have been published yet. Furthermore, there<br />
was a lack <strong>of</strong> test harnesses <strong>and</strong> benchmarks using representative workloads for EBS. Consequently,<br />
we focused on the development <strong>of</strong> a performance modeling methodology <strong>of</strong> EBS as well<br />
as on approaches to benchmark them. We summarize now our main contributions <strong>and</strong> proposed<br />
approaches.<br />
To comprehend our contributions, an underst<strong>and</strong>ing <strong>of</strong> the fundamental ideas <strong>of</strong> EBS is essential.<br />
Therefore, we discussed our underst<strong>and</strong>ing <strong>of</strong> events in detail <strong>and</strong> introduced definitions<br />
for different kinds <strong>of</strong> events <strong>and</strong> related concepts. Generally spoken, EBS are s<strong>of</strong>tware systems<br />
in which an observed event triggers a reaction. We evaluated the variety <strong>of</strong> underlying technologies<br />
with a focus on DEBS <strong>and</strong> MOMs <strong>and</strong> provided a survey <strong>of</strong> existing DEBS <strong>and</strong> MOM<br />
products <strong>and</strong> st<strong>and</strong>ards. In our review <strong>of</strong> existing work, we identified a lack <strong>of</strong> benchmarks <strong>and</strong><br />
performance modeling approaches for EBS. To support a structural evaluation <strong>of</strong> benchmarks,<br />
we introduced five categories <strong>of</strong> requirements: (i) Representativeness: the benchmark has to be<br />
based on a representative workload. (ii) Comprehensiveness: exercise all platform features typically<br />
used in applications. (iii)Focus: place the emphasis on the technology server <strong>and</strong> minimize<br />
the impact <strong>of</strong> other services, e.g., databases. (iv) Configurability: provide a configurable tool for<br />
performance analysis. (v) Scalability: provide ways to scale the workload in a flexible manner.<br />
None <strong>of</strong> the existing benchmarks met all our requirements. Therefore, we saw a strong need<br />
for independent <strong>and</strong> st<strong>and</strong>ardized benchmarks for EBS fulfilling the requirements. We launched<br />
a project inside the SPEC with the goal to develop the first industry st<strong>and</strong>ard benchmark for<br />
EBS. As underlying technology plattform we chose JMS. This was motivated by the fact that<br />
MOMs are widely used in industry <strong>and</strong> the quasi-st<strong>and</strong>ard for MOMs is JMS. Under the lead<br />
<strong>of</strong> TU Darmstadt a team with members from organizations such as IBM, Sun <strong>and</strong> BEA was<br />
formed. Our efforts resulted in the SPECjms2007 st<strong>and</strong>ard benchmark. Its main contributions<br />
were tw<strong>of</strong>old:<br />
St<strong>and</strong>ard Workload: <strong>Based</strong> on the feedback <strong>of</strong> our industrial partners, we specified a comprehensive<br />
workload with different scaling options that fullfils all our requirements. It contains<br />
135